How do you get the prompt for unit test generation

[zb1439]

Thanks for sharing the work! I wonder how did you choose the code solution for the unit test generation prompt in data/benchmark/input_humaneval+_ut.jsonl, is it a randomly selected solution from previous execution?

BTW, do you have plans to share the data preparation and/or SFT code for reproduction? Thanks!

[zb1439]

In addition, how do you get the xx_anno.jsonl file at https://github.com/RUCKBReasoning/CodeRM/blob/main/evaluation/calculate_result.py#L74C9-L74C89. It seems like none of the steps in the reproduction section of README.md could generate a jsonl file that generates the required task_sol_results which is not based on generated unit tests but also has result as part of the field of a json struct.